Merge branch 'omap-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tmlind...
[linux-2.6] / arch / s390 / kernel / smp.c
1 /*
2  *  arch/s390/kernel/smp.c
3  *
4  *    Copyright IBM Corp. 1999,2007
5  *    Author(s): Denis Joseph Barrow (djbarrow@de.ibm.com,barrow_dj@yahoo.com),
6  *               Martin Schwidefsky (schwidefsky@de.ibm.com)
7  *               Heiko Carstens (heiko.carstens@de.ibm.com)
8  *
9  *  based on other smp stuff by
10  *    (c) 1995 Alan Cox, CymruNET Ltd  <alan@cymru.net>
11  *    (c) 1998 Ingo Molnar
12  *
13  * We work with logical cpu numbering everywhere we can. The only
14  * functions using the real cpu address (got from STAP) are the sigp
15  * functions. For all other functions we use the identity mapping.
16  * That means that cpu_number_map[i] == i for every cpu. cpu_number_map is
17  * used e.g. to find the idle task belonging to a logical cpu. Every array
18  * in the kernel is sorted by the logical cpu number and not by the physical
19  * one which is causing all the confusion with __cpu_logical_map and
20  * cpu_number_map in other architectures.
21  */
22
23 #include <linux/module.h>
24 #include <linux/init.h>
25 #include <linux/mm.h>
26 #include <linux/err.h>
27 #include <linux/spinlock.h>
28 #include <linux/kernel_stat.h>
29 #include <linux/delay.h>
30 #include <linux/cache.h>
31 #include <linux/interrupt.h>
32 #include <linux/cpu.h>
33 #include <linux/timex.h>
34 #include <linux/bootmem.h>
35 #include <asm/ipl.h>
36 #include <asm/setup.h>
37 #include <asm/sigp.h>
38 #include <asm/pgalloc.h>
39 #include <asm/irq.h>
40 #include <asm/s390_ext.h>
41 #include <asm/cpcmd.h>
42 #include <asm/tlbflush.h>
43 #include <asm/timer.h>
44 #include <asm/lowcore.h>
45 #include <asm/sclp.h>
46 #include <asm/cpu.h>
47 #include "entry.h"
48
49 /*
50  * An array with a pointer the lowcore of every CPU.
51  */
52 struct _lowcore *lowcore_ptr[NR_CPUS];
53 EXPORT_SYMBOL(lowcore_ptr);
54
55 cpumask_t cpu_online_map = CPU_MASK_NONE;
56 EXPORT_SYMBOL(cpu_online_map);
57
58 cpumask_t cpu_possible_map = CPU_MASK_ALL;
59 EXPORT_SYMBOL(cpu_possible_map);
60
61 static struct task_struct *current_set[NR_CPUS];
62
63 static u8 smp_cpu_type;
64 static int smp_use_sigp_detection;
65
66 enum s390_cpu_state {
67         CPU_STATE_STANDBY,
68         CPU_STATE_CONFIGURED,
69 };
70
71 DEFINE_MUTEX(smp_cpu_state_mutex);
72 int smp_cpu_polarization[NR_CPUS];
73 static int smp_cpu_state[NR_CPUS];
74 static int cpu_management;
75
76 static DEFINE_PER_CPU(struct cpu, cpu_devices);
77
78 static void smp_ext_bitcall(int, ec_bit_sig);
79
80 /*
81  * Structure and data for __smp_call_function_map(). This is designed to
82  * minimise static memory requirements. It also looks cleaner.
83  */
84 static DEFINE_SPINLOCK(call_lock);
85
86 struct call_data_struct {
87         void (*func) (void *info);
88         void *info;
89         cpumask_t started;
90         cpumask_t finished;
91         int wait;
92 };
93
94 static struct call_data_struct *call_data;
95
96 /*
97  * 'Call function' interrupt callback
98  */
99 static void do_call_function(void)
100 {
101         void (*func) (void *info) = call_data->func;
102         void *info = call_data->info;
103         int wait = call_data->wait;
104
105         cpu_set(smp_processor_id(), call_data->started);
106         (*func)(info);
107         if (wait)
108                 cpu_set(smp_processor_id(), call_data->finished);;
109 }
110
111 static void __smp_call_function_map(void (*func) (void *info), void *info,
112                                     int nonatomic, int wait, cpumask_t map)
113 {
114         struct call_data_struct data;
115         int cpu, local = 0;
116
117         /*
118          * Can deadlock when interrupts are disabled or if in wrong context.
119          */
120         WARN_ON(irqs_disabled() || in_irq());
121
122         /*
123          * Check for local function call. We have to have the same call order
124          * as in on_each_cpu() because of machine_restart_smp().
125          */
126         if (cpu_isset(smp_processor_id(), map)) {
127                 local = 1;
128                 cpu_clear(smp_processor_id(), map);
129         }
130
131         cpus_and(map, map, cpu_online_map);
132         if (cpus_empty(map))
133                 goto out;
134
135         data.func = func;
136         data.info = info;
137         data.started = CPU_MASK_NONE;
138         data.wait = wait;
139         if (wait)
140                 data.finished = CPU_MASK_NONE;
141
142         call_data = &data;
143
144         for_each_cpu_mask(cpu, map)
145                 smp_ext_bitcall(cpu, ec_call_function);
146
147         /* Wait for response */
148         while (!cpus_equal(map, data.started))
149                 cpu_relax();
150         if (wait)
151                 while (!cpus_equal(map, data.finished))
152                         cpu_relax();
153 out:
154         if (local) {
155                 local_irq_disable();
156                 func(info);
157                 local_irq_enable();
158         }
159 }
160
161 /*
162  * smp_call_function:
163  * @func: the function to run; this must be fast and non-blocking
164  * @info: an arbitrary pointer to pass to the function
165  * @nonatomic: unused
166  * @wait: if true, wait (atomically) until function has completed on other CPUs
167  *
168  * Run a function on all other CPUs.
169  *
170  * You must not call this function with disabled interrupts, from a
171  * hardware interrupt handler or from a bottom half.
172  */
173 int smp_call_function(void (*func) (void *info), void *info, int nonatomic,
174                       int wait)
175 {
176         cpumask_t map;
177
178         spin_lock(&call_lock);
179         map = cpu_online_map;
180         cpu_clear(smp_processor_id(), map);
181         __smp_call_function_map(func, info, nonatomic, wait, map);
182         spin_unlock(&call_lock);
183         return 0;
184 }
185 EXPORT_SYMBOL(smp_call_function);
186
187 /*
188  * smp_call_function_single:
189  * @cpu: the CPU where func should run
190  * @func: the function to run; this must be fast and non-blocking
191  * @info: an arbitrary pointer to pass to the function
192  * @nonatomic: unused
193  * @wait: if true, wait (atomically) until function has completed on other CPUs
194  *
195  * Run a function on one processor.
196  *
197  * You must not call this function with disabled interrupts, from a
198  * hardware interrupt handler or from a bottom half.
199  */
200 int smp_call_function_single(int cpu, void (*func) (void *info), void *info,
201                              int nonatomic, int wait)
202 {
203         spin_lock(&call_lock);
204         __smp_call_function_map(func, info, nonatomic, wait,
205                                 cpumask_of_cpu(cpu));
206         spin_unlock(&call_lock);
207         return 0;
208 }
209 EXPORT_SYMBOL(smp_call_function_single);
210
211 /**
212  * smp_call_function_mask(): Run a function on a set of other CPUs.
213  * @mask: The set of cpus to run on.  Must not include the current cpu.
214  * @func: The function to run. This must be fast and non-blocking.
215  * @info: An arbitrary pointer to pass to the function.
216  * @wait: If true, wait (atomically) until function has completed on other CPUs.
217  *
218  * Returns 0 on success, else a negative status code.
219  *
220  * If @wait is true, then returns once @func has returned; otherwise
221  * it returns just before the target cpu calls @func.
222  *
223  * You must not call this function with disabled interrupts or from a
224  * hardware interrupt handler or from a bottom half handler.
225  */
226 int smp_call_function_mask(cpumask_t mask, void (*func)(void *), void *info,
227                            int wait)
228 {
229         spin_lock(&call_lock);
230         cpu_clear(smp_processor_id(), mask);
231         __smp_call_function_map(func, info, 0, wait, mask);
232         spin_unlock(&call_lock);
233         return 0;
234 }
235 EXPORT_SYMBOL(smp_call_function_mask);
236
237 void smp_send_stop(void)
238 {
239         int cpu, rc;
240
241         /* Disable all interrupts/machine checks */
242         __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
243
244         /* write magic number to zero page (absolute 0) */
245         lowcore_ptr[smp_processor_id()]->panic_magic = __PANIC_MAGIC;
246
247         /* stop all processors */
248         for_each_online_cpu(cpu) {
249                 if (cpu == smp_processor_id())
250                         continue;
251                 do {
252                         rc = signal_processor(cpu, sigp_stop);
253                 } while (rc == sigp_busy);
254
255                 while (!smp_cpu_not_running(cpu))
256                         cpu_relax();
257         }
258 }
259
260 /*
261  * This is the main routine where commands issued by other
262  * cpus are handled.
263  */
264
265 static void do_ext_call_interrupt(__u16 code)
266 {
267         unsigned long bits;
268
269         /*
270          * handle bit signal external calls
271          *
272          * For the ec_schedule signal we have to do nothing. All the work
273          * is done automatically when we return from the interrupt.
274          */
275         bits = xchg(&S390_lowcore.ext_call_fast, 0);
276
277         if (test_bit(ec_call_function, &bits))
278                 do_call_function();
279 }
280
281 /*
282  * Send an external call sigp to another cpu and return without waiting
283  * for its completion.
284  */
285 static void smp_ext_bitcall(int cpu, ec_bit_sig sig)
286 {
287         /*
288          * Set signaling bit in lowcore of target cpu and kick it
289          */
290         set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
291         while (signal_processor(cpu, sigp_emergency_signal) == sigp_busy)
292                 udelay(10);
293 }
294
295 #ifndef CONFIG_64BIT
296 /*
297  * this function sends a 'purge tlb' signal to another CPU.
298  */
299 static void smp_ptlb_callback(void *info)
300 {
301         __tlb_flush_local();
302 }
303
304 void smp_ptlb_all(void)
305 {
306         on_each_cpu(smp_ptlb_callback, NULL, 0, 1);
307 }
308 EXPORT_SYMBOL(smp_ptlb_all);
309 #endif /* ! CONFIG_64BIT */
310
311 /*
312  * this function sends a 'reschedule' IPI to another CPU.
313  * it goes straight through and wastes no time serializing
314  * anything. Worst case is that we lose a reschedule ...
315  */
316 void smp_send_reschedule(int cpu)
317 {
318         smp_ext_bitcall(cpu, ec_schedule);
319 }
320
321 /*
322  * parameter area for the set/clear control bit callbacks
323  */
324 struct ec_creg_mask_parms {
325         unsigned long orvals[16];
326         unsigned long andvals[16];
327 };
328
329 /*
330  * callback for setting/clearing control bits
331  */
332 static void smp_ctl_bit_callback(void *info)
333 {
334         struct ec_creg_mask_parms *pp = info;
335         unsigned long cregs[16];
336         int i;
337
338         __ctl_store(cregs, 0, 15);
339         for (i = 0; i <= 15; i++)
340                 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
341         __ctl_load(cregs, 0, 15);
342 }
343
344 /*
345  * Set a bit in a control register of all cpus
346  */
347 void smp_ctl_set_bit(int cr, int bit)
348 {
349         struct ec_creg_mask_parms parms;
350
351         memset(&parms.orvals, 0, sizeof(parms.orvals));
352         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
353         parms.orvals[cr] = 1 << bit;
354         on_each_cpu(smp_ctl_bit_callback, &parms, 0, 1);
355 }
356 EXPORT_SYMBOL(smp_ctl_set_bit);
357
358 /*
359  * Clear a bit in a control register of all cpus
360  */
361 void smp_ctl_clear_bit(int cr, int bit)
362 {
363         struct ec_creg_mask_parms parms;
364
365         memset(&parms.orvals, 0, sizeof(parms.orvals));
366         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
367         parms.andvals[cr] = ~(1L << bit);
368         on_each_cpu(smp_ctl_bit_callback, &parms, 0, 1);
369 }
370 EXPORT_SYMBOL(smp_ctl_clear_bit);
371
372 /*
373  * In early ipl state a temp. logically cpu number is needed, so the sigp
374  * functions can be used to sense other cpus. Since NR_CPUS is >= 2 on
375  * CONFIG_SMP and the ipl cpu is logical cpu 0, it must be 1.
376  */
377 #define CPU_INIT_NO     1
378
379 #if defined(CONFIG_ZFCPDUMP) || defined(CONFIG_ZFCPDUMP_MODULE)
380
381 /*
382  * zfcpdump_prefix_array holds prefix registers for the following scenario:
383  * 64 bit zfcpdump kernel and 31 bit kernel which is to be dumped. We have to
384  * save its prefix registers, since they get lost, when switching from 31 bit
385  * to 64 bit.
386  */
387 unsigned int zfcpdump_prefix_array[NR_CPUS + 1] \
388         __attribute__((__section__(".data")));
389
390 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
391 {
392         if (ipl_info.type != IPL_TYPE_FCP_DUMP)
393                 return;
394         if (cpu >= NR_CPUS) {
395                 printk(KERN_WARNING "Registers for cpu %i not saved since dump "
396                        "kernel was compiled with NR_CPUS=%i\n", cpu, NR_CPUS);
397                 return;
398         }
399         zfcpdump_save_areas[cpu] = kmalloc(sizeof(union save_area), GFP_KERNEL);
400         __cpu_logical_map[CPU_INIT_NO] = (__u16) phy_cpu;
401         while (signal_processor(CPU_INIT_NO, sigp_stop_and_store_status) ==
402                sigp_busy)
403                 cpu_relax();
404         memcpy(zfcpdump_save_areas[cpu],
405                (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
406                SAVE_AREA_SIZE);
407 #ifdef CONFIG_64BIT
408         /* copy original prefix register */
409         zfcpdump_save_areas[cpu]->s390x.pref_reg = zfcpdump_prefix_array[cpu];
410 #endif
411 }
412
413 union save_area *zfcpdump_save_areas[NR_CPUS + 1];
414 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
415
416 #else
417
418 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
419
420 #endif /* CONFIG_ZFCPDUMP || CONFIG_ZFCPDUMP_MODULE */
421
422 static int cpu_stopped(int cpu)
423 {
424         __u32 status;
425
426         /* Check for stopped state */
427         if (signal_processor_ps(&status, 0, cpu, sigp_sense) ==
428             sigp_status_stored) {
429                 if (status & 0x40)
430                         return 1;
431         }
432         return 0;
433 }
434
435 static int cpu_known(int cpu_id)
436 {
437         int cpu;
438
439         for_each_present_cpu(cpu) {
440                 if (__cpu_logical_map[cpu] == cpu_id)
441                         return 1;
442         }
443         return 0;
444 }
445
446 static int smp_rescan_cpus_sigp(cpumask_t avail)
447 {
448         int cpu_id, logical_cpu;
449
450         logical_cpu = first_cpu(avail);
451         if (logical_cpu == NR_CPUS)
452                 return 0;
453         for (cpu_id = 0; cpu_id <= 65535; cpu_id++) {
454                 if (cpu_known(cpu_id))
455                         continue;
456                 __cpu_logical_map[logical_cpu] = cpu_id;
457                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
458                 if (!cpu_stopped(logical_cpu))
459                         continue;
460                 cpu_set(logical_cpu, cpu_present_map);
461                 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
462                 logical_cpu = next_cpu(logical_cpu, avail);
463                 if (logical_cpu == NR_CPUS)
464                         break;
465         }
466         return 0;
467 }
468
469 static int smp_rescan_cpus_sclp(cpumask_t avail)
470 {
471         struct sclp_cpu_info *info;
472         int cpu_id, logical_cpu, cpu;
473         int rc;
474
475         logical_cpu = first_cpu(avail);
476         if (logical_cpu == NR_CPUS)
477                 return 0;
478         info = kmalloc(sizeof(*info), GFP_KERNEL);
479         if (!info)
480                 return -ENOMEM;
481         rc = sclp_get_cpu_info(info);
482         if (rc)
483                 goto out;
484         for (cpu = 0; cpu < info->combined; cpu++) {
485                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
486                         continue;
487                 cpu_id = info->cpu[cpu].address;
488                 if (cpu_known(cpu_id))
489                         continue;
490                 __cpu_logical_map[logical_cpu] = cpu_id;
491                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
492                 cpu_set(logical_cpu, cpu_present_map);
493                 if (cpu >= info->configured)
494                         smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
495                 else
496                         smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
497                 logical_cpu = next_cpu(logical_cpu, avail);
498                 if (logical_cpu == NR_CPUS)
499                         break;
500         }
501 out:
502         kfree(info);
503         return rc;
504 }
505
506 static int __smp_rescan_cpus(void)
507 {
508         cpumask_t avail;
509
510         cpus_xor(avail, cpu_possible_map, cpu_present_map);
511         if (smp_use_sigp_detection)
512                 return smp_rescan_cpus_sigp(avail);
513         else
514                 return smp_rescan_cpus_sclp(avail);
515 }
516
517 static void __init smp_detect_cpus(void)
518 {
519         unsigned int cpu, c_cpus, s_cpus;
520         struct sclp_cpu_info *info;
521         u16 boot_cpu_addr, cpu_addr;
522
523         c_cpus = 1;
524         s_cpus = 0;
525         boot_cpu_addr = S390_lowcore.cpu_data.cpu_addr;
526         info = kmalloc(sizeof(*info), GFP_KERNEL);
527         if (!info)
528                 panic("smp_detect_cpus failed to allocate memory\n");
529         /* Use sigp detection algorithm if sclp doesn't work. */
530         if (sclp_get_cpu_info(info)) {
531                 smp_use_sigp_detection = 1;
532                 for (cpu = 0; cpu <= 65535; cpu++) {
533                         if (cpu == boot_cpu_addr)
534                                 continue;
535                         __cpu_logical_map[CPU_INIT_NO] = cpu;
536                         if (!cpu_stopped(CPU_INIT_NO))
537                                 continue;
538                         smp_get_save_area(c_cpus, cpu);
539                         c_cpus++;
540                 }
541                 goto out;
542         }
543
544         if (info->has_cpu_type) {
545                 for (cpu = 0; cpu < info->combined; cpu++) {
546                         if (info->cpu[cpu].address == boot_cpu_addr) {
547                                 smp_cpu_type = info->cpu[cpu].type;
548                                 break;
549                         }
550                 }
551         }
552
553         for (cpu = 0; cpu < info->combined; cpu++) {
554                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
555                         continue;
556                 cpu_addr = info->cpu[cpu].address;
557                 if (cpu_addr == boot_cpu_addr)
558                         continue;
559                 __cpu_logical_map[CPU_INIT_NO] = cpu_addr;
560                 if (!cpu_stopped(CPU_INIT_NO)) {
561                         s_cpus++;
562                         continue;
563                 }
564                 smp_get_save_area(c_cpus, cpu_addr);
565                 c_cpus++;
566         }
567 out:
568         kfree(info);
569         printk(KERN_INFO "CPUs: %d configured, %d standby\n", c_cpus, s_cpus);
570         get_online_cpus();
571         __smp_rescan_cpus();
572         put_online_cpus();
573 }
574
575 /*
576  *      Activate a secondary processor.
577  */
578 int __cpuinit start_secondary(void *cpuvoid)
579 {
580         /* Setup the cpu */
581         cpu_init();
582         preempt_disable();
583         /* Enable TOD clock interrupts on the secondary cpu. */
584         init_cpu_timer();
585 #ifdef CONFIG_VIRT_TIMER
586         /* Enable cpu timer interrupts on the secondary cpu. */
587         init_cpu_vtimer();
588 #endif
589         /* Enable pfault pseudo page faults on this cpu. */
590         pfault_init();
591
592         /* Mark this cpu as online */
593         spin_lock(&call_lock);
594         cpu_set(smp_processor_id(), cpu_online_map);
595         spin_unlock(&call_lock);
596         /* Switch on interrupts */
597         local_irq_enable();
598         /* Print info about this processor */
599         print_cpu_info(&S390_lowcore.cpu_data);
600         /* cpu_idle will call schedule for us */
601         cpu_idle();
602         return 0;
603 }
604
605 static void __init smp_create_idle(unsigned int cpu)
606 {
607         struct task_struct *p;
608
609         /*
610          *  don't care about the psw and regs settings since we'll never
611          *  reschedule the forked task.
612          */
613         p = fork_idle(cpu);
614         if (IS_ERR(p))
615                 panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
616         current_set[cpu] = p;
617         spin_lock_init(&(&per_cpu(s390_idle, cpu))->lock);
618 }
619
620 static int __cpuinit smp_alloc_lowcore(int cpu)
621 {
622         unsigned long async_stack, panic_stack;
623         struct _lowcore *lowcore;
624         int lc_order;
625
626         lc_order = sizeof(long) == 8 ? 1 : 0;
627         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, lc_order);
628         if (!lowcore)
629                 return -ENOMEM;
630         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
631         panic_stack = __get_free_page(GFP_KERNEL);
632         if (!panic_stack || !async_stack)
633                 goto out;
634         memcpy(lowcore, &S390_lowcore, 512);
635         memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
636         lowcore->async_stack = async_stack + ASYNC_SIZE;
637         lowcore->panic_stack = panic_stack + PAGE_SIZE;
638
639 #ifndef CONFIG_64BIT
640         if (MACHINE_HAS_IEEE) {
641                 unsigned long save_area;
642
643                 save_area = get_zeroed_page(GFP_KERNEL);
644                 if (!save_area)
645                         goto out_save_area;
646                 lowcore->extended_save_area_addr = (u32) save_area;
647         }
648 #endif
649         lowcore_ptr[cpu] = lowcore;
650         return 0;
651
652 #ifndef CONFIG_64BIT
653 out_save_area:
654         free_page(panic_stack);
655 #endif
656 out:
657         free_pages(async_stack, ASYNC_ORDER);
658         free_pages((unsigned long) lowcore, lc_order);
659         return -ENOMEM;
660 }
661
662 #ifdef CONFIG_HOTPLUG_CPU
663 static void smp_free_lowcore(int cpu)
664 {
665         struct _lowcore *lowcore;
666         int lc_order;
667
668         lc_order = sizeof(long) == 8 ? 1 : 0;
669         lowcore = lowcore_ptr[cpu];
670 #ifndef CONFIG_64BIT
671         if (MACHINE_HAS_IEEE)
672                 free_page((unsigned long) lowcore->extended_save_area_addr);
673 #endif
674         free_page(lowcore->panic_stack - PAGE_SIZE);
675         free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
676         free_pages((unsigned long) lowcore, lc_order);
677         lowcore_ptr[cpu] = NULL;
678 }
679 #endif /* CONFIG_HOTPLUG_CPU */
680
681 /* Upping and downing of CPUs */
682 int __cpuinit __cpu_up(unsigned int cpu)
683 {
684         struct task_struct *idle;
685         struct _lowcore *cpu_lowcore;
686         struct stack_frame *sf;
687         sigp_ccode ccode;
688
689         if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
690                 return -EIO;
691         if (smp_alloc_lowcore(cpu))
692                 return -ENOMEM;
693
694         ccode = signal_processor_p((__u32)(unsigned long)(lowcore_ptr[cpu]),
695                                    cpu, sigp_set_prefix);
696         if (ccode) {
697                 printk("sigp_set_prefix failed for cpu %d "
698                        "with condition code %d\n",
699                        (int) cpu, (int) ccode);
700                 return -EIO;
701         }
702
703         idle = current_set[cpu];
704         cpu_lowcore = lowcore_ptr[cpu];
705         cpu_lowcore->kernel_stack = (unsigned long)
706                 task_stack_page(idle) + THREAD_SIZE;
707         cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
708         sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
709                                      - sizeof(struct pt_regs)
710                                      - sizeof(struct stack_frame));
711         memset(sf, 0, sizeof(struct stack_frame));
712         sf->gprs[9] = (unsigned long) sf;
713         cpu_lowcore->save_area[15] = (unsigned long) sf;
714         __ctl_store(cpu_lowcore->cregs_save_area, 0, 15);
715         asm volatile(
716                 "       stam    0,15,0(%0)"
717                 : : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
718         cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
719         cpu_lowcore->current_task = (unsigned long) idle;
720         cpu_lowcore->cpu_data.cpu_nr = cpu;
721         cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce;
722         cpu_lowcore->ipl_device = S390_lowcore.ipl_device;
723         eieio();
724
725         while (signal_processor(cpu, sigp_restart) == sigp_busy)
726                 udelay(10);
727
728         while (!cpu_online(cpu))
729                 cpu_relax();
730         return 0;
731 }
732
733 static int __init setup_possible_cpus(char *s)
734 {
735         int pcpus, cpu;
736
737         pcpus = simple_strtoul(s, NULL, 0);
738         cpu_possible_map = cpumask_of_cpu(0);
739         for (cpu = 1; cpu < pcpus && cpu < NR_CPUS; cpu++)
740                 cpu_set(cpu, cpu_possible_map);
741         return 0;
742 }
743 early_param("possible_cpus", setup_possible_cpus);
744
745 #ifdef CONFIG_HOTPLUG_CPU
746
747 int __cpu_disable(void)
748 {
749         struct ec_creg_mask_parms cr_parms;
750         int cpu = smp_processor_id();
751
752         cpu_clear(cpu, cpu_online_map);
753
754         /* Disable pfault pseudo page faults on this cpu. */
755         pfault_fini();
756
757         memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
758         memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
759
760         /* disable all external interrupts */
761         cr_parms.orvals[0] = 0;
762         cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 12 |
763                                 1 << 11 | 1 << 10 | 1 <<  6 | 1 <<  4);
764         /* disable all I/O interrupts */
765         cr_parms.orvals[6] = 0;
766         cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
767                                 1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
768         /* disable most machine checks */
769         cr_parms.orvals[14] = 0;
770         cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
771                                  1 << 25 | 1 << 24);
772
773         smp_ctl_bit_callback(&cr_parms);
774
775         return 0;
776 }
777
778 void __cpu_die(unsigned int cpu)
779 {
780         /* Wait until target cpu is down */
781         while (!smp_cpu_not_running(cpu))
782                 cpu_relax();
783         smp_free_lowcore(cpu);
784         printk(KERN_INFO "Processor %d spun down\n", cpu);
785 }
786
787 void cpu_die(void)
788 {
789         idle_task_exit();
790         signal_processor(smp_processor_id(), sigp_stop);
791         BUG();
792         for (;;);
793 }
794
795 #endif /* CONFIG_HOTPLUG_CPU */
796
797 void __init smp_prepare_cpus(unsigned int max_cpus)
798 {
799 #ifndef CONFIG_64BIT
800         unsigned long save_area = 0;
801 #endif
802         unsigned long async_stack, panic_stack;
803         struct _lowcore *lowcore;
804         unsigned int cpu;
805         int lc_order;
806
807         smp_detect_cpus();
808
809         /* request the 0x1201 emergency signal external interrupt */
810         if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
811                 panic("Couldn't request external interrupt 0x1201");
812         print_cpu_info(&S390_lowcore.cpu_data);
813
814         /* Reallocate current lowcore, but keep its contents. */
815         lc_order = sizeof(long) == 8 ? 1 : 0;
816         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, lc_order);
817         panic_stack = __get_free_page(GFP_KERNEL);
818         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
819 #ifndef CONFIG_64BIT
820         if (MACHINE_HAS_IEEE)
821                 save_area = get_zeroed_page(GFP_KERNEL);
822 #endif
823         local_irq_disable();
824         local_mcck_disable();
825         lowcore_ptr[smp_processor_id()] = lowcore;
826         *lowcore = S390_lowcore;
827         lowcore->panic_stack = panic_stack + PAGE_SIZE;
828         lowcore->async_stack = async_stack + ASYNC_SIZE;
829 #ifndef CONFIG_64BIT
830         if (MACHINE_HAS_IEEE)
831                 lowcore->extended_save_area_addr = (u32) save_area;
832 #endif
833         set_prefix((u32)(unsigned long) lowcore);
834         local_mcck_enable();
835         local_irq_enable();
836         for_each_possible_cpu(cpu)
837                 if (cpu != smp_processor_id())
838                         smp_create_idle(cpu);
839 }
840
841 void __init smp_prepare_boot_cpu(void)
842 {
843         BUG_ON(smp_processor_id() != 0);
844
845         current_thread_info()->cpu = 0;
846         cpu_set(0, cpu_present_map);
847         cpu_set(0, cpu_online_map);
848         S390_lowcore.percpu_offset = __per_cpu_offset[0];
849         current_set[0] = current;
850         smp_cpu_state[0] = CPU_STATE_CONFIGURED;
851         smp_cpu_polarization[0] = POLARIZATION_UNKNWN;
852         spin_lock_init(&(&__get_cpu_var(s390_idle))->lock);
853 }
854
855 void __init smp_cpus_done(unsigned int max_cpus)
856 {
857 }
858
859 /*
860  * the frequency of the profiling timer can be changed
861  * by writing a multiplier value into /proc/profile.
862  *
863  * usually you want to run this on all CPUs ;)
864  */
865 int setup_profiling_timer(unsigned int multiplier)
866 {
867         return 0;
868 }
869
870 #ifdef CONFIG_HOTPLUG_CPU
871 static ssize_t cpu_configure_show(struct sys_device *dev, char *buf)
872 {
873         ssize_t count;
874
875         mutex_lock(&smp_cpu_state_mutex);
876         count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
877         mutex_unlock(&smp_cpu_state_mutex);
878         return count;
879 }
880
881 static ssize_t cpu_configure_store(struct sys_device *dev, const char *buf,
882                                    size_t count)
883 {
884         int cpu = dev->id;
885         int val, rc;
886         char delim;
887
888         if (sscanf(buf, "%d %c", &val, &delim) != 1)
889                 return -EINVAL;
890         if (val != 0 && val != 1)
891                 return -EINVAL;
892
893         get_online_cpus();
894         mutex_lock(&smp_cpu_state_mutex);
895         rc = -EBUSY;
896         if (cpu_online(cpu))
897                 goto out;
898         rc = 0;
899         switch (val) {
900         case 0:
901                 if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
902                         rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
903                         if (!rc) {
904                                 smp_cpu_state[cpu] = CPU_STATE_STANDBY;
905                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
906                         }
907                 }
908                 break;
909         case 1:
910                 if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
911                         rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
912                         if (!rc) {
913                                 smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
914                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
915                         }
916                 }
917                 break;
918         default:
919                 break;
920         }
921 out:
922         mutex_unlock(&smp_cpu_state_mutex);
923         put_online_cpus();
924         return rc ? rc : count;
925 }
926 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
927 #endif /* CONFIG_HOTPLUG_CPU */
928
929 static ssize_t cpu_polarization_show(struct sys_device *dev, char *buf)
930 {
931         int cpu = dev->id;
932         ssize_t count;
933
934         mutex_lock(&smp_cpu_state_mutex);
935         switch (smp_cpu_polarization[cpu]) {
936         case POLARIZATION_HRZ:
937                 count = sprintf(buf, "horizontal\n");
938                 break;
939         case POLARIZATION_VL:
940                 count = sprintf(buf, "vertical:low\n");
941                 break;
942         case POLARIZATION_VM:
943                 count = sprintf(buf, "vertical:medium\n");
944                 break;
945         case POLARIZATION_VH:
946                 count = sprintf(buf, "vertical:high\n");
947                 break;
948         default:
949                 count = sprintf(buf, "unknown\n");
950                 break;
951         }
952         mutex_unlock(&smp_cpu_state_mutex);
953         return count;
954 }
955 static SYSDEV_ATTR(polarization, 0444, cpu_polarization_show, NULL);
956
957 static ssize_t show_cpu_address(struct sys_device *dev, char *buf)
958 {
959         return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
960 }
961 static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
962
963
964 static struct attribute *cpu_common_attrs[] = {
965 #ifdef CONFIG_HOTPLUG_CPU
966         &attr_configure.attr,
967 #endif
968         &attr_address.attr,
969         &attr_polarization.attr,
970         NULL,
971 };
972
973 static struct attribute_group cpu_common_attr_group = {
974         .attrs = cpu_common_attrs,
975 };
976
977 static ssize_t show_capability(struct sys_device *dev, char *buf)
978 {
979         unsigned int capability;
980         int rc;
981
982         rc = get_cpu_capability(&capability);
983         if (rc)
984                 return rc;
985         return sprintf(buf, "%u\n", capability);
986 }
987 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
988
989 static ssize_t show_idle_count(struct sys_device *dev, char *buf)
990 {
991         struct s390_idle_data *idle;
992         unsigned long long idle_count;
993
994         idle = &per_cpu(s390_idle, dev->id);
995         spin_lock_irq(&idle->lock);
996         idle_count = idle->idle_count;
997         spin_unlock_irq(&idle->lock);
998         return sprintf(buf, "%llu\n", idle_count);
999 }
1000 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
1001
1002 static ssize_t show_idle_time(struct sys_device *dev, char *buf)
1003 {
1004         struct s390_idle_data *idle;
1005         unsigned long long new_time;
1006
1007         idle = &per_cpu(s390_idle, dev->id);
1008         spin_lock_irq(&idle->lock);
1009         if (idle->in_idle) {
1010                 new_time = get_clock();
1011                 idle->idle_time += new_time - idle->idle_enter;
1012                 idle->idle_enter = new_time;
1013         }
1014         new_time = idle->idle_time;
1015         spin_unlock_irq(&idle->lock);
1016         return sprintf(buf, "%llu\n", new_time >> 12);
1017 }
1018 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
1019
1020 static struct attribute *cpu_online_attrs[] = {
1021         &attr_capability.attr,
1022         &attr_idle_count.attr,
1023         &attr_idle_time_us.attr,
1024         NULL,
1025 };
1026
1027 static struct attribute_group cpu_online_attr_group = {
1028         .attrs = cpu_online_attrs,
1029 };
1030
1031 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
1032                                     unsigned long action, void *hcpu)
1033 {
1034         unsigned int cpu = (unsigned int)(long)hcpu;
1035         struct cpu *c = &per_cpu(cpu_devices, cpu);
1036         struct sys_device *s = &c->sysdev;
1037         struct s390_idle_data *idle;
1038
1039         switch (action) {
1040         case CPU_ONLINE:
1041         case CPU_ONLINE_FROZEN:
1042                 idle = &per_cpu(s390_idle, cpu);
1043                 spin_lock_irq(&idle->lock);
1044                 idle->idle_enter = 0;
1045                 idle->idle_time = 0;
1046                 idle->idle_count = 0;
1047                 spin_unlock_irq(&idle->lock);
1048                 if (sysfs_create_group(&s->kobj, &cpu_online_attr_group))
1049                         return NOTIFY_BAD;
1050                 break;
1051         case CPU_DEAD:
1052         case CPU_DEAD_FROZEN:
1053                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1054                 break;
1055         }
1056         return NOTIFY_OK;
1057 }
1058
1059 static struct notifier_block __cpuinitdata smp_cpu_nb = {
1060         .notifier_call = smp_cpu_notify,
1061 };
1062
1063 static int __devinit smp_add_present_cpu(int cpu)
1064 {
1065         struct cpu *c = &per_cpu(cpu_devices, cpu);
1066         struct sys_device *s = &c->sysdev;
1067         int rc;
1068
1069         c->hotpluggable = 1;
1070         rc = register_cpu(c, cpu);
1071         if (rc)
1072                 goto out;
1073         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1074         if (rc)
1075                 goto out_cpu;
1076         if (!cpu_online(cpu))
1077                 goto out;
1078         rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1079         if (!rc)
1080                 return 0;
1081         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1082 out_cpu:
1083 #ifdef CONFIG_HOTPLUG_CPU
1084         unregister_cpu(c);
1085 #endif
1086 out:
1087         return rc;
1088 }
1089
1090 #ifdef CONFIG_HOTPLUG_CPU
1091
1092 int __ref smp_rescan_cpus(void)
1093 {
1094         cpumask_t newcpus;
1095         int cpu;
1096         int rc;
1097
1098         get_online_cpus();
1099         mutex_lock(&smp_cpu_state_mutex);
1100         newcpus = cpu_present_map;
1101         rc = __smp_rescan_cpus();
1102         if (rc)
1103                 goto out;
1104         cpus_andnot(newcpus, cpu_present_map, newcpus);
1105         for_each_cpu_mask(cpu, newcpus) {
1106                 rc = smp_add_present_cpu(cpu);
1107                 if (rc)
1108                         cpu_clear(cpu, cpu_present_map);
1109         }
1110         rc = 0;
1111 out:
1112         mutex_unlock(&smp_cpu_state_mutex);
1113         put_online_cpus();
1114         if (!cpus_empty(newcpus))
1115                 topology_schedule_update();
1116         return rc;
1117 }
1118
1119 static ssize_t __ref rescan_store(struct sys_device *dev, const char *buf,
1120                                   size_t count)
1121 {
1122         int rc;
1123
1124         rc = smp_rescan_cpus();
1125         return rc ? rc : count;
1126 }
1127 static SYSDEV_ATTR(rescan, 0200, NULL, rescan_store);
1128 #endif /* CONFIG_HOTPLUG_CPU */
1129
1130 static ssize_t dispatching_show(struct sys_device *dev, char *buf)
1131 {
1132         ssize_t count;
1133
1134         mutex_lock(&smp_cpu_state_mutex);
1135         count = sprintf(buf, "%d\n", cpu_management);
1136         mutex_unlock(&smp_cpu_state_mutex);
1137         return count;
1138 }
1139
1140 static ssize_t dispatching_store(struct sys_device *dev, const char *buf,
1141                                  size_t count)
1142 {
1143         int val, rc;
1144         char delim;
1145
1146         if (sscanf(buf, "%d %c", &val, &delim) != 1)
1147                 return -EINVAL;
1148         if (val != 0 && val != 1)
1149                 return -EINVAL;
1150         rc = 0;
1151         get_online_cpus();
1152         mutex_lock(&smp_cpu_state_mutex);
1153         if (cpu_management == val)
1154                 goto out;
1155         rc = topology_set_cpu_management(val);
1156         if (!rc)
1157                 cpu_management = val;
1158 out:
1159         mutex_unlock(&smp_cpu_state_mutex);
1160         put_online_cpus();
1161         return rc ? rc : count;
1162 }
1163 static SYSDEV_ATTR(dispatching, 0644, dispatching_show, dispatching_store);
1164
1165 static int __init topology_init(void)
1166 {
1167         int cpu;
1168         int rc;
1169
1170         register_cpu_notifier(&smp_cpu_nb);
1171
1172 #ifdef CONFIG_HOTPLUG_CPU
1173         rc = sysfs_create_file(&cpu_sysdev_class.kset.kobj,
1174                                &attr_rescan.attr);
1175         if (rc)
1176                 return rc;
1177 #endif
1178         rc = sysfs_create_file(&cpu_sysdev_class.kset.kobj,
1179                                &attr_dispatching.attr);
1180         if (rc)
1181                 return rc;
1182         for_each_present_cpu(cpu) {
1183                 rc = smp_add_present_cpu(cpu);
1184                 if (rc)
1185                         return rc;
1186         }
1187         return 0;
1188 }
1189 subsys_initcall(topology_init);